Improve string comparison diffs with a linear-space edit-distance algorithm

Spock’s EditDistance currently stores the complete Levenshtein matrix, requiring O(N * M) memory as well as O(N * M) time.
The protection introduced for #121 prevents OutOfMemoryError, but it does so by limiting the compared region. When the differing section exceeds the limit, failed string comparisons only report:
Strings too large to calculate edit distance.
This means moderately large strings can lose their useful inline diff even when the calculation would be practical with a more memory-efficient algorithm.
Replace the dense matrix with a linear-space optimal-alignment algorithm, such as the Myers-Miller/Hirschberg divide-and-conquer algorithm.

Note: The existing EditDistance Javadoc explicitly lists using an algorithm with lower time and/or space complexity as a potential improvement.
id est

/*
 * Ideas for improvements:
 * - Favor fewer EditOperationS when calculating distance and/or path
 * - Use algorithm with lower time and/or space complexity
 */

This issue tracks addressing that TODO by replacing the dense matrix with a linear-space optimal-alignment algorithm.

Start by locating Spock’s EditDistance implementation and its Javadoc, then inspect how the current dense Levenshtein matrix and size protection produce comparison diffs. Replace that approach with a linear-space optimal-alignment algorithm while preserving useful inline diffs for larger strings. Done means the dense matrix is no longer required and the existing comparison behavior remains correct.